Sun glint, the specular reflection of sunlight off the ocean surface and into the satellite sensor, compromises data processing in ocean remote sensing imagery. Several correcting algorithms and models have been developed to improve satellite high-level products quality in glint-contaminated areas. Some rely on radiative transfer equations and others on in-scene information analysis, but most of these methods are based on an estimation of the sun glint contribution in terms of radiance. In this paper, we introduce a new approach for sun glint study in medium resolution satellite images, derived from multiscale/multifractal techniques. Since sun glint introduces long range effects and occurs at a scale much smaller than pixel dimensions, multiscale techniques are well suited to analyse such complex signals and are able to extract features related to the oceanic flows. First, we apply a multifractal decomposition scheme to generate the signal singularity exponents and retrieve the so-called "most singular manifold". Using the main streamlines of the signal, we compute a reduced image containing the same multiscale structures as the original one but a naive distribution of the gradient. We will then show that the correlation between the original image and the reduced one, also known as the source field in the microcanonical multiscale formalism, displays a signature related to the presence of sun glint.